Power Aware Parallel 3-D Finite Element Mesh Refinement Performance Modeling and Analysis With CUDA/MPI on GPU and Multi-Core Architecture

Abstract

Software power performance tuning handles the critical design constraints of software running on hardware platforms composed of large numbers of power-hungry components. The power dissipation of a Single Program/Instruction Multiple Data (SPMD/SIMD) computation such as finite element method (FEM) mesh refinement is highly dependent on the underlying algorithm and the power-consuming features of hardware Processing Elements (PE). This contribution presents a practical methodology for modeling and analyzing the power performance of parallel 3-D FEM mesh refinement on CUDA/MPI architecture based on detailed software prototypes and power parameters in order to predict the power functionality and runtime behavior of the algorithm, optimize the program design and thus achieve the best power efficiency. In detail, we have proposed approaches for GPU parallelization, dynamic CPU frequency scaling and dynamic load scheduling among PEs. The performance improvement of our designs has been demonstrated and the results have been validated on a real multi-core and GPU cluster.

Exploring Mesh Generation and Quality Enhancement with Open Source Codes

“…These values are important, mainly considering the 3D nontrivial domains, which were chosen to test our proposal. Therefore, unless the parallel computing resources are used (Bracken et al, 2012), neither the topological domain complexity, nor the mesh refinements and time processing performance of the examples discussed in relevant studies (Li et al, 2013;Chen and Biro, 2012;Zhao et al, 2012;Ho et al, 2011;Zhang and Kumar, 2011;Krebs et al, 2010) surpass the equivalent features of the test cases presented in this study, which were explored using one CPU core. Moreover, the approach described is able to provide highly quality meshes with common and very accessible hardware resources.…”

“…The proposed method was tested considering solids with complex geometries: "C-type" magnet, turbocompressor device, Klein bottle and an electrical motor with supplementary permanent magnets. These choices were motivated by the several studies focused on computational electromagnetics considering similar geometries (Li et al, 2016;Chen and Biro, 2012;Wall et al, 2012;Zhao et al, 2012;Bracken et al, 2012;Zhang and Kumar, 2011;Ho et al, 2011;Chang et al, 2010;Jang et al, 2007;Cho et al, 2006).…”

“…All meshes included in this study do not represent the refinement limits, which could be reached, even using one core of the processor. In this context, several methods have been developed to study mesh generation and FEM simulations (Bracken et al, 2012;Li et al, 2013;Chen and Biro, 2012;Zhao et al, 2012;Ho et al, 2011;Zhang and Kumar, 2011;Krebs et al, 2010). However, none of them has used our proposed combination.…”

“…One of the reasons for this, is that the more complex is the geometry of the problem domain, the more sophisticated are the software tools for geometric modeling and mesh discretization required by the FEM. In this context, several papers have been addressed the topic of mesh generation, for both 2 and 3dimensional elements, in different areas (Cacace and Blöcher, 2015;Leng et al, 2013;Wang and Yu, 2012;Chen and Biro, 2012;Wall et al, 2012;Zhao et al, 2012; 1001 Bracken et al, 2012;Ecabert et al, 2011;Zhang and Kumar, 2011;Ho et al, 2011;Krebs et al, 2010;Sun and Chaichana, 2010;Milasinovic et al, 2008;Barauskas et al, 2007). In these studies, strategies of algorithms and software tools for domain construction and mesh generation, well as the performance and the availability of the tools were present issues.…”

Mapping Cohesive Fracture and Fragmentation Simulations to Graphics Processor Units

“…Kakay et al implement their finite element micromagnetic simulation in the GPU, demonstrating a high speed performance compared to CPU multi‐core implementation. Ren et al model and analyze power performance of parallel 3D finite element mesh refinement on CUDA and Message Passing Interface (MPI) architecture using multi‐core CPU and GPU cluster, also proposing parallelization techniques for both. In recent works on FEM, Markall et al use finite element advection‐diffusion solver to demonstrate that FEM implementations on many‐core (GPU) and multi‐core (CPU) architectures differ if their performance potential is to be obtained.…”